Home > Archive > 2018 > Volume 8 Number 6 (Dec. 2018) >
IJMLC 2018 Vol.8(6): 606-612 ISSN: 2010-3700
DOI: 10.18178/ijmlc.2018.8.6.753

Optimizing ANN Training Performance for Chaotic Time Series Prediction Using Small Data Size

Lei Zhang

Abstract—In this paper, the training performance of artificial neural network (ANN) is investigated based on three aspects of the training data: the data size, the distribution of the subsets for training, validation and testing, and the data segments of the training data, which are generated using Lorenz chaotic system equations and the forward Euler method to represent the chaotic features commonly found in real world applications, specifically for the Electroencephalogram (EEG) signals captured from brain activities. This research investigates the potentiality and feasibility of using small data set for training ANN to generate chaotic time series, which can be used for the simulation and analysis of chaotic features in EEG signals. Contradictory to the popular belief that better performance can be achieved by larger number of training samples, the training results show that the same level of training performance can be achieved by a relative small number of training samples for the generation and prediction of the chaotic system time series.

Index Terms—Electroencephalogram (EEG), time series, artificial neural network (ANN), small data, chaotic patterns.

L. Zhang is with the Faculty of Engineering and Applied Science, University of Regina, Regina, SK, Canada, S4S 6T2 (e-mail: lei.zhang@uregina.ca).

[PDF]

Cite: Lei Zhang, "Optimizing ANN Training Performance for Chaotic Time Series Prediction Using Small Data Size," International Journal of Machine Learning and Computing vol. 8, no. 6, pp. 606-612, 2018.

General Information

  • ISSN: 2010-3700 (Online)
  • Abbreviated Title: Int. J. Mach. Learn. Comput.
  • Frequency: Bimonthly
  • DOI: 10.18178/IJMLC
  • Editor-in-Chief: Dr. Lin Huang
  • Executive Editor:  Ms. Cherry L. Chen
  • Abstracing/Indexing: Inspec (IET), Google Scholar, Crossref, ProQuest, Electronic Journals Library.
  • E-mail: ijmlc@ejournal.net


Article Metrics